Compiled String File format

Supplied by xezon

Header

4 bytes: ' FSC': file type identifier
4 bytes: 3: unknown value, maybe file type version
4 bytes: 0x1916: 6422, could be number of string table entries
4 bytes: 0x1915: 6421, similar to number above, not clear what the difference is
8 bytes: all 0: unknown

Body (repeats)

4 bytes: ' LBL': keyword identifier
4 bytes: 1: unknown value, maybe number of keywords? seems to be 1 always.
4 bytes: length of ascii string (N)
N bytes: ascii string
4 bytes: ' RTS': locale string identifier
4 bytes: length of unicode string (N)
N bytes: unicode string: XOR'ed by 0xFF

Supplied by Thyme Wiki

Source: https://github.com/TheAssemblyArmada/Thyme/wiki/Compiled-String-File-Format

This file format holds most of the game strings in an encoded format that decodes to UCS-2 Unicode strings. The file consists of a header followed by a series of ASCII label strings and encoded Unicode strings.

Header

enum LanguageID : int32_t
{
    LANGUAGE_ID_US,
    LANGUAGE_ID_UK,
    LANGUAGE_ID_GERMAN,
    LANGUAGE_ID_FRENCH,
    LANGUAGE_ID_SPANISH,
    LANGUAGE_ID_ITALIAN,
    LANGUAGE_ID_JAPANSE,
    LANGUAGE_ID_JABBER,
    LANGUAGE_ID_KOREAN,
    LANGUAGE_ID_CHINESE,
    LANGUAGE_ID_UNK1,
    LANGUAGE_ID_UNK2,
    LANGUAGE_ID_POLISH,
    LANGUAGE_ID_UNKNOWN,
};

struct CSFHeader
{
    uint32_t id;
    int32_t version;
    int32_t num_labels;
    int32_t num_strings;
    int32_t skip;
    LanguageID langid;
};

The id is a FourCC code that translates to the ASCII string “ FSC”, essentially “CSF “ in little-endian format.

version is always 3 for SAGE engine games, though the engine does check for a value of 1 or less, in which case the langid is ignored and set to 0.

num_labels is a count of how many string references there are in the file, while num_strings is a count of how many encoded strings there are as each label can have more than one, though this feature is unused in SAGE games.

skip is a reserved 4 bytes that are unused.

langid corresponds to an internal enum that identifies which language the file is intended to provide text for, though the SAGE engine games don’t appear to make use of it. The provided enum is valid for Generals and Zero Hour, but not the Battle for Middle Earth games based on examination of the string files from different translations.

Labels

struct Label
{
    uint32_t id;
    int32_t string_count;
    int32_t length;
    char label[length]
};

Each label entry begins with a FourCC id equivalent to the string “ LBL”, again likely a little-endian “LBL “. This is followed by the number of strings the label can refer to, then the length of the label ASCII string followed by the string itself. The label string is not null-terminated and must match the given length. The engine code refers only to the label when requesting a string for display so that localisation can be done separately.

Strings

struct String
{
    uint32_t id;
    int32_t length;
    uint16_t string[length]
    int32_t ex_length;         // Only present when id matches 'WRTS'
    char ex_string[ex_length]; // Only present when id matches 'WRTS'
};

The FourCC id for the string must be either “ RTS” or WRTS (little-endian “STR “ and STRW), with WRTS indicating an additional ASCII string appended on. The values for the string are little endian and are encoded using a binary NOT on the value. The following is an example of how it might be decoded in C:

for (int i = 0; i < length; ++i) {
    string[i] = ~string[i];
}

Debug information left in certain executables suggests that ex_string was related to audio files in some way, possibly as a file name for an audio file containing a reading of the string text. However it appears such a feature is not used in any SAGE game.